<html dir="ltr">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style id="owaParaStyle" type="text/css">P {margin-top:0;margin-bottom:0;}</style>
</head>
<body ocsi="0" fpstyle="1">
<div style="direction: ltr;font-family: Tahoma;color: #000000;font-size: 10pt;">Hi guys,<br>
<br>
Thanks for getting on top of this so quickly. I'm looking forward to seeing your solution to the problem.
<br>
<br>
@Xiaoming: Nice job with the example you provided! I'll send you a copy of my statfun later.<br>
<br>
Best,<br>
Martin<br>
<br>
<div style="font-family: Times New Roman; color: #000000; font-size: 16px">
<hr tabindex="-1">
<div style="direction: ltr;" id="divRpF14706"><font face="Tahoma" color="#000000" size="2"><b>Von:</b> fieldtrip-bounces@science.ru.nl [fieldtrip-bounces@science.ru.nl]" im Auftrag von "Arjen Stolk [a.stolk8@gmail.com]<br>
<b>Gesendet:</b> Donnerstag, 22. Oktober 2015 06:01<br>
<b>An:</b> FieldTrip discussion list<br>
<b>Betreff:</b> Re: [FieldTrip] calculating behavioural-power correlation -- follow-up questions<br>
</font><br>
</div>
<div></div>
<div>
<div dir="ltr">Hi Xiaoming, Martin,
<div><br>
</div>
<div>Coming back about this issue, I have filed a bug report:</div>
<div><a href="http://bugzilla.fieldtriptoolbox.org/show_bug.cgi?id=2992" target="_blank">http://bugzilla.fieldtriptoolbox.org/show_bug.cgi?id=2992</a><br>
</div>
<div><br>
</div>
<div>In short, creating a randomization distribution using the permutation method is prone to systematic bias across conditions, as illustrated in our previous email exchanges. One workaround is to use the bootstrapping method (cfg.resampling = 'bootstrap'),
or to explicitly remove the systematic bias, e.g by normalizing the data in each condition separately prior to doing statistical testing.</div>
<div><br>
</div>
<div>I have added these important considerations to the documentation of ft_statfun_correlationT, built in a warning message, and created a latent function that explicitly tests for this issue (to prevent it to re-occur with future code changes). Thank you
for your acuteness and contributions.</div>
<div><br>
</div>
<div>Yours,</div>
<div>Arjen</div>
<div><br>
</div>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">2015-10-21 15:33 GMT-07:00 Arjen Stolk <span dir="ltr"><<a href="mailto:a.stolk8@gmail.com" target="_blank">a.stolk8@gmail.com</a>></span>:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex; border-left:1px #ccc solid; padding-left:1ex">
<div dir="ltr">Thanks for clarifying, Xiaoming.
<div><br>
</div>
<div>It seems indeed that a systematic offset (in averages) across the two sample populations biases the randomization distribution. I'd need to sort this out, but schematically would expect it to look as follow:</div>
<div><br>
</div>
<div style="font-size:13px">normal:</div>
<div style="font-size:13px">~1000 ~1000 ~1000 ~1000</div>
<div style="font-size:13px">~1 ~1 ~1 ~1</div>
<div style="font-size:13px">corr = ~0</div>
<div style="font-size:13px"><br>
</div>
<div style="font-size:13px">post shuffling:</div>
<div style="font-size:13px">
<div>~1000 ~1 ~1000 ~1000</div>
<div>~1 ~1000 ~1 ~1</div>
</div>
<div><span style="font-size:13px">corr = negative, that is an anti-correlation</span></div>
<div><br>
</div>
<div>This implies a bootstrapping approach is more correct for testing for statistical significance of a correlation (in case of a systematic offset across conditions).<br>
</div>
<div><br>
</div>
<div>Yours,</div>
<div>Arjen</div>
</div>
<div class="HOEnZb">
<div class="h5">
<div class="gmail_extra"><br>
<div class="gmail_quote">2015-10-21 14:28 GMT-07:00 Xiaoming Du <span dir="ltr"><<a href="mailto:XDu@mprc.umaryland.edu" target="_blank">XDu@mprc.umaryland.edu</a>></span>:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex; border-left:1px #ccc solid; padding-left:1ex">
<div style="margin:4px 4px 1px; font:10pt Tahoma">
<div>Hi Martin,</div>
<div> </div>
<div>Thanks Martin to make it more clear. However, my understanding is that the problem is caused by different means of permutation for paired-sample t test and correlation. I am very curious how you solve the problem by customizing statfun. Do you mind sending
me a copy of that? I will really appreciate it. </div>
<div> </div>
<div>Hi Arjen,</div>
<div> </div>
<div>I think the 'ft_statfun_correlationT' works fine. Using the code in your previous email, the correlation result (stat.rho) is same as using MATLAB function 'corr' with Spearman method. However, In your code, the design matrix for correlation is defined
in the same way as for within-UO design (e.g., paired-sample t test). I think Martin's and my concern is that the permutation method for correlation and within-UO design (e.g., paired-sample t test) should be different.
</div>
<div> </div>
<div>For example, there are 10 subjects (patients). Each of them finished a task before (A) and after (B) the treatment . Based on the different null hypothesis, the permutation methods for paired-sample t test and correlation should be different. For paired-sample
t test (or within-UO design), the null hypothesis is that there is no different before or after the treatment. In other words, for each subjects, switch the results before and after treatment won't matter. This is exactly how the permutations are done when
cfg.design is defined as in your code . On the other hand, the null hypothesis for correlation is that: there is no relationship between the results before and after treatment. Therefore, permutation should be done on after-treatment results
<u>among subjects</u> while keep before-treatment results same (or permute before-treatment and keep after-treatment). In terms of permutation, correlation and paired-sample t test (within-UO design) seems treated equally in fieldtrip.
</div>
<div> <br>
Original matrix permutation for paired t test permutation for correlation</div>
<div> </div>
<div> before (A) after (B) before (A) after (B) before (A) after (B)
<br>
subj1 A1 B1 subj1 A1 B1 subj1 A1 B2<br>
subj2 A2 B2 subj2 B2 A2 subj2 A2 B4<br>
subj3 A3 B3 subj3 B3 A3 subj3 A3 B3<br>
subj4 A4 B4 subj4 A4 B4 subj4 A4 B1<br>
subj5 A5 B5 subj5 A5 B5 subj5 A5 B10<br>
subj6 A6 B6 subj6 B6 A6 subj6 A6 B8<br>
subj7 A7 B7 subj7 B7 A7 subj7 A7 B9<br>
subj8 A8 B8 subj8 A8 B8 subj8 A8 B6<br>
subj9 A9 B9 subj9 B9 A9 subj9 A9 B5<br>
subj10 A10 B10 subj10 A10 B10 subj10 A10 B7</div>
<div> </div>
<div>As follows, I made minor changes in your previous code to show this issue. </div>
<div>data_brain ranges from 0 to 1;</div>
<div>data_behav ranges from 1000 to 1001;</div>
<div>There is no relationship between data_brain and data_behav, because the values are randomly assigned to them.</div>
<div>This is the results I got by using the code below: </div>
<div> </div>
<div>stat = </div>
<div> </div>
<div> prob: 9.9900e-04<br>
cirange: 0.0020<br>
mask: 1<br>
stat: 0.2941<br>
ref: -10.7655<br>
rho: 0.0297<br>
dimord: 'chan_time'<br>
label: {'1'}<br>
time: 1<br>
cfg: [1x1 struct]</div>
<div> </div>
<div>stat.rho (0.0297) is the same as using 'corr' with Spearman option in matlab; However, stat.prob obtained by using permutation test is so small. The reason (I guess) are 1, two-tail test was used. 2, the rho distribution of permutation is biased to negative
direction (meaning the rho after each permutation was biased to strong negative correlation. e.g., -0.8 etc ), so the real rho fell into the 5% range of rho distribution of permutations. In other words, real rho (0.0297) is smaller (or larger) than almost
all rho from permutation, so that stat.prob is very small (0.000999). </div>
<div> </div>
<div>Please let me know if this makes sense to you or if I misunderstood anything.
</div>
<div> </div>
<div>Thanks.</div>
<div> </div>
<div>Xiaoming</div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div>data_brain = [];</div>
<div> </div>
<div>data_behav = [];</div>
<div> </div>
<div>for j=1:100</div>
<div> </div>
<div>data_brain{j}.avg = rand; % increasing </div>
<span>
<div> </div>
<div>data_brain{j}.dimord = 'chan_time';</div>
<div> </div>
<div>data_brain{j}.time = 1;</div>
<div> </div>
<div>data_brain{j}.label = {'1'};</div>
<div> </div>
<div><br>
data_behav{j} = data_brain{j};</div>
<div> </div>
</span>
<div>data_behav{j}.avg = rand + 1000; % add scaling difference</div>
<span>
<div> </div>
<div>end</div>
<div> </div>
<div><br>
% compute statistics with correlationT</div>
<div> </div>
<div>cfg = [];</div>
<div> </div>
<div>cfg.method = 'montecarlo';</div>
<div> </div>
<div>cfg.statistic = 'ft_statfun_correlationT';</div>
<div> </div>
</span>
<div>cfg.numrandomization = 1000;</div>
<div> </div>
<div>n1 = 100; % n1 is the number of subjects</div>
<span>
<div> </div>
<div>design = zeros(2, n1 * 2);</div>
<div> </div>
<div>design(1,1:n1) = 1;</div>
<div> </div>
<div>design(1,(n1 + 1):(n1 * 2)) = 2;</div>
<div> </div>
<div>design(2, :) = [1:n1 1:n1];</div>
<div> </div>
<div>cfg.design = design;</div>
<div> </div>
<div><br>
cfg.ivar = 1;</div>
<div> </div>
<div>cfg.uvar = 2;</div>
<div> </div>
<div>stat = ft_timelockstatistics(cfg, data_brain{:}, data_behav{:});<br>
</div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
<div> </div>
</span>
<div><br>
<br>
>>> Arjen Stolk <<a href="mailto:a.stolk8@gmail.com" target="_blank">a.stolk8@gmail.com</a>> 10/21/2015 2:22 PM >>><br>
</div>
<div>
<div>
<div dir="ltr">Hi Martin,
<div><br>
</div>
<div>Thanks for thinking along. <span style="font-size:13px">I've briefly tried to replicate/simplify the situation depicted by you using the code below. It works as I would expect, no matter whether one variable is scaled differently. But perhaps I'm not fully
capturing the issue, and something still goes awry. This is a possibility because ft_statfun_correlationT has only recently been implemented for a specific case, and was never really tested within different situations (hence it's not well-documented on the
wiki). Do you think you could use this example code to replicate the situation you are experiencing?</span></div>
<div>
<div style="font-size:13px"><br>
</div>
<div style="font-size:13px">Yours,</div>
<div style="font-size:13px">Arjen
<div><br>
</div>
<div>
<p style="line-height:normal; margin:0px; font-family:Courier; color:rgb(37,153,45); font-size:10px">
% simulate simple multiple subjects timelock structures</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_brain = [];</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_behav = [];</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px"><span style="color:rgb(4,51,255)">for</span> j=1:10</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_brain{j}.avg = j;
<span style="color:rgb(37,153,45)">% increasing </span></p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_brain{j}.dimord =
<span style="color:rgb(178,69,243)">'chan_time'</span>;</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_brain{j}.time = 1;</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_brain{j}.label = {<span style="color:rgb(178,69,243)">'1'</span>};</p>
<p style="line-height:normal; margin:0px; min-height:12px; font-family:Courier; font-size:10px">
</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_behav{j} = data_brain{j};</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">data_behav{j}.avg = data_brain{j}.avg*-1000+50;
<span style="color:rgb(37,153,45)">% add scaling difference</span></p>
<p style="line-height:normal; margin:0px; font-family:Courier; color:rgb(4,51,255); font-size:10px">
end</p>
<p style="line-height:normal; margin:0px; min-height:12px; font-family:Courier; color:rgb(4,51,255); font-size:10px">
</p>
<p style="line-height:normal; margin:0px; font-family:Courier; color:rgb(37,153,45); font-size:10px">
% compute statistics with correlationT</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">cfg = [];</p>
<span>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">cfg.method =
<span style="color:rgb(178,69,243)">'montecarlo'</span>;</p>
<p style="line-height:normal; margin:0px; font-family:Courier; color:rgb(178,69,243); font-size:10px">
<span style="color:rgb(0,0,0)">cfg.statistic = </span>'ft_statfun_correlationT'<span style="color:rgb(0,0,0)">;</span></p>
</span>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">cfg.numrandomization = 100;</p>
<p style="line-height:normal; margin:0px; min-height:12px; font-family:Courier; font-size:10px">
</p>
<p style="line-height:normal; margin:0px; font-family:Courier; color:rgb(37,153,45); font-size:10px">
<span style="color:rgb(0,0,0)">n1 = 10; </span>% n1 is the number of subjects</p>
<span>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">design = zeros(2, n1 * 2);</p>
</span>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">design(1,1:n1) = 1;</p>
<span>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">design(1,(n1 + 1):(n1 * 2)) = 2;</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">design(2, :) = [1:n1 1:n1];</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">cfg.design = design;</p>
<p style="line-height:normal; margin:0px; min-height:12px; font-family:Courier; font-size:10px">
</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">cfg.ivar = 1;</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">cfg.uvar = 2;</p>
</span>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">stat = ft_timelockstatistics(cfg, data_brain{:}, data_behav{:});</p>
<p style="line-height:normal; margin:0px; min-height:12px; font-family:Courier; font-size:10px">
</p>
<p style="line-height:normal; margin:0px; font-family:Courier; font-size:10px">assert(isequal(stat.rho, -1));</p>
</div>
</div>
</div>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">2015-10-21 10:54 GMT-07:00 Krebber, Martin <span dir="ltr">
<<a href="mailto:martin.krebber@charite.de" target="_blank">martin.krebber@charite.de</a>></span>:<br>
<blockquote class="gmail_quote" style="border-left:#ccc 1px solid; margin:0px 0px 0px 0.8ex; padding-left:1ex">
<div>
<div style="font-family:Tahoma; direction:ltr; color:#000000; font-size:10pt">Hi Xiaoming, hi Arjen,<br>
<br>
I've been encountering the same problem. I believe Xiaoming is right when he points out that the permutaion step shuffles data across conditions and that this introduces a negative bias in the distribution. I found the same thing when I correlated RT data with
TFRs (absolute power). My distribution was shifted strongly to the left and, thus, not a single negative cluster was significant, but every positive one was.
<br>
<br>
Xiaomings explanation made a lot of sense to me when I thought about it graphically: Imagine correlating two data vectors, one (x) ranging between .5 and1, the other (y) between 50 and and 100. When plotting this, one gets a cloud of dots on the upper left
corner of the diagram. When you then switch the variable assignment of half of the data points (which is what the permutation step seems to do), these dots will now be be shifted to the lower right corner of the diagram. So no matter what the correlation in
the original data, chances are that (given different scaling) after permutaion, you get a negative correlation.<br>
<br>
I am not 100% sure about this, so please let me know if I made a mistake.<br>
<br>
What I tried instead of the 'ft_statfun_correlationT' was using a custom made statfun in which I pass the RTs via the design matrix. With this, my results looked much better. I am not sure, but I guess this is because there is no shuffling between the two variables
in this case.<br>
<br>
I would really like to know, what is the right way of doing this using just the FieldTrip functions. Is there a way to permute data within variables? I tried cfg.resampling = 'bootstrap', but this is not a permutation, as far as I know.<br>
<br>
<br>
Thanks!<br>
Martin<br>
<br>
<br>
<div style="font-family:Times New Roman; color:#000000; font-size:16px">
<hr>
<div style="direction:ltr"><font face="Tahoma" color="#000000" size="2"><b>Von:</b>
<a href="mailto:fieldtrip-bounces@science.ru.nl" target="_blank">fieldtrip-bounces@science.ru.nl</a> [<a href="mailto:fieldtrip-bounces@science.ru.nl" target="_blank">fieldtrip-bounces@science.ru.nl</a>]" im Auftrag von "Arjen Stolk [<a href="mailto:a.stolk8@gmail.com" target="_blank">a.stolk8@gmail.com</a>]<br>
<b>Gesendet:</b> Dienstag, 20. Oktober 2015 08:03<br>
<b>An:</b> FieldTrip discussion list<br>
<b>Betreff:</b> Re: [FieldTrip] calculating behavioural-power correlation -- follow-up questions<br>
</font><br>
</div>
<div>
<div>
<div></div>
<div>
<div dir="ltr">Hey Xiaoming,
<div><br>
</div>
<div>It's still pretty hard, for me, to guess on basis of that matlab output what is going on here and what you mean with 'shuffling design matrices', and how that shuffling 'biases the cluster distribution'. As you mention yourself, it could be due to various
reasons, and you're open to suggestions and increasing your understanding. I'd therefore suggest to try to funnel the number of potential explanations by simulating what you're doing (using input data for which you know how it should behave), after you've
read more about what the design matrix and monte carlo statistics are supposed to do. Perhaps the statistics section at the bottom of this page provides a good starting point:
<a href="http://www.fieldtriptoolbox.org/walkthrough" target="_blank">http://www.fieldtriptoolbox.org/walkthrough</a></div>
<div><br>
</div>
<div>Hope that helps,</div>
<div>Arjen </div>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">2015-10-19 15:56 GMT-07:00 Xiaoming Du <span dir="ltr"><<a href="mailto:XDu@mprc.umaryland.edu" target="_blank">XDu@mprc.umaryland.edu</a>></span>:<br>
<blockquote class="gmail_quote" style="border-left:#ccc 1px solid; margin:0px 0px 0px 0.8ex; padding-left:1ex">
<div style="margin:4px 4px 1px; font:10pt Tahoma">
<div>
<div>
<div>For example, our power values ranged from 1 to 3 (after log transform); my behavioral data ranged from 20 to 90;</div>
<div></div>
<div>by using above mentioned script, there are 14 negative clusters were reported in variable stat.</div>
<div></div>
<div>stat = </div>
<div></div>
<div>prob: [30x50 double]<br>
posclusters: []<br>
posclusterslabelmat: [30x50 double]<br>
posdistribution: [1x1000 double]<br>
negclusters: [1x14 struct]<br>
negclusterslabelmat: [30x50 double]<br>
negdistribution: [1x1000 double]<br>
cirange: [30x50 double]<br>
mask: [30x50 logical]<br>
stat: [30x50 double]<br>
ref: [30x50 double]<br>
rho: [30x50 double]<br>
dimord: 'chan_freq'<br>
freq: [1x50 double]<br>
label: {30x1 cell}<br>
time: 2.5000<br>
cfg: [1x1 struct]</div>
<div></div>
<div>However, the p values of those clusters (i.e., stat.negclusters.prob) are all ones. The smallest value in stat.negdistribution is way larger than the largest negative cluster t-sum. This could be real. However, it is more likely due to the shuffle between
power and behavioral group. For example, design matrix [1 1 1 1 2 2 2 2; 1 2 3 4 1 2 3 4] was shuffled to [1 2 2 1 2 2 1 1; 1 2 3 4 1 2 3 4]. After each permutation, for some subjects, their power data was labeled as behavioral data and vice versa. Because
of the scale difference between power and behavioral data, large negative correlations were generated by permutation. This further biased the cluster distribution.
</div>
<div>My limited understanding is that, for correlation, each permutation should fix cfg.ivar and only shuffle half of the cfg.uvar. For example, permute design matrix [1 1 1 1 2 2 2 2; 1 2 3 4 1 2 3 4] to [1 1 1 1 2 2 2; 1 2 3 4 4 2 3 1]. THerefore, after permutation,
one subject's power data corresponds to another subject's behavioral data.</div>
<div></div>
<div>I am not good at statistics. It will be really appreciated if you have any suggestions or comments.</div>
<div></div>
<div>Xiaoming</div>
<div></div>
<div></div>
<div><br>
<br>
>>> Arjen Stolk <<a href="mailto:a.stolk8@gmail.com" target="_blank">a.stolk8@gmail.com</a>> 10/19/2015 6:01 PM >>><br>
</div>
</div>
</div>
<div>
<div>
<div dir="ltr">Hey Xiaoming,
<div><br>
</div>
<div>Not sure if I understand, but shouldn't the directions of the correlations be independent of the scaling of the two variables? Looking at the code of ft_statfun_correlationT it doesn't seem the conversion from correlation to T value (tstat = rho*(sqrt(max(nunits)-2))/sqrt((1-rho^2)))
would result in a direction change either. Perhaps you could try to first manually calculate a correlation between signal power and behavioral power, and see whether anything is behaving unexpectedly?
</div>
<div><br>
</div>
<div>Yours,</div>
<div>Arjen</div>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">2015-10-19 14:25 GMT-07:00 Xiaoming Du <span dir="ltr"><<a href="mailto:XDu@mprc.umaryland.edu" target="_blank">XDu@mprc.umaryland.edu</a>></span>:<br>
<blockquote class="gmail_quote" style="border-left:#ccc 1px solid; margin:0px 0px 0px 0.8ex; padding-left:1ex">
<div style="margin:4px 4px 1px; font:10pt Tahoma">
<div>Dear FieldTrip users,</div>
<div></div>
<div>This is Xiaoming from University of Maryland Baltimore. My current project requires to calculate behavioral-power correlation across subjects. Similar topic was discussed here early this year.
<a href="http://mailman.science.ru.nl/pipermail/fieldtrip/2015-February/008953.html" target="_blank">
http://mailman.science.ru.nl/pipermail/fieldtrip/2015-February/008953.html</a></div>
<div></div>
<div>According to the suggestions in above mentioned thread, I duplicate my power dataset and replace the power values at each time-frequency point with behavioral data. Therefore, those two datasets have same structure and dimension. I used the following script
to test if there are significant clusters of correlations.</div>
<div></div>
<div>cfg = [];<br>
cfg.parameter = 'powspctrm';<br>
cfg.method = 'montecarlo';<br>
cfg.statistic = 'ft_statfun_correlationT';<br>
...</div>
<div>etc</div>
<div>...<br>
design = zeros(2, n1 * 2); % n1 is the number of subjects.<br>
design(1,1:n1) = 1;<br>
design(1,(n1 + 1):(n1 * 2)) = 2;<br>
design(2, :) = [[1:n1 ] [1 : n1]];<br>
cfg.design = design;</div>
<div><br>
cfg.ivar = 1;<br>
cfg.uvar = 2;<br>
stat = ft_freqstatistics(cfg, dataBeh{:}, dataDX1{:});</div>
<div></div>
<div></div>
<div>However, it seems when each time the design matrix is permuted, FieldTrip is using the same method as for 'ft_statfun_depsamplesT', meaning cfg.uvar remains the same while cfg.ivar (1 or 2) is randomly assigned to each subject in design matrix. Although
I confirmed this by uncommenting line 313 (i.e., tmpdesign = design(:,resample(i,:))) in ft_statistics_montecarlo.m which allows to display the permuted design matrix in command line, please correct me if this is not the case.
</div>
<div></div>
<div>In my mind, this kind of permutation will cause trouble when dealing with correlation. For example, in my case, the behavioral data and power data have different scales. The power data are much larger than behavioral data in general. When assigning behavioral
data into power group or vice versa, it will induce huge negative correlations between power and behavioral measurement. Therefore, no negative clusters will survive from permutation test.
</div>
<div></div>
<div>Please let me know if I have mis-understanding or if I did anything wrong. Any suggestions will be highly appreciated!</div>
<div></div>
<div>Thanks.</div>
<span><font color="#888888">
<div></div>
<div>Xiaoming </div>
<div></div>
<div></div>
<div></div>
<div></div>
<div></div>
</font></span></div>
<br>
_______________________________________________<br>
fieldtrip mailing list<br>
<a href="mailto:fieldtrip@donders.ru.nl" target="_blank">fieldtrip@donders.ru.nl</a><br>
<a href="http://mailman.science.ru.nl/mailman/listinfo/fieldtrip" rel="noreferrer" target="_blank">http://mailman.science.ru.nl/mailman/listinfo/fieldtrip</a><br>
</blockquote>
</div>
<br>
</div>
</div>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</div>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</div>
</div>
</div>
</body>
</html>